Flexible Data Centers Aim to Alleviate Electric Grid Strain
A new approach seeks to integrate data centers more flexibly with existing electric grids, aiming to manage peak power demands. Emerald AI's software, Conductor, has been developed to enable data centers to reduce their power consumption during high-demand periods, preventing potential blackouts or grid instability. This innovation, tested in simulations and set for live deployment with partners like Nvidia and Digital Realty, addresses the challenge of long lead times for new power plant construction and public concerns over data centers' energy footprint. By adapting to grid needs, these 'power-flexible AI factories' could accelerate data center deployment and optimize energy use within current infrastructure.
Data centers, known for their significant power consumption, are exploring methods to operate more flexibly within existing electric grids. This initiative aims to address the growing demand for electricity while circumventing the slow pace of new power plant development.
In December 2025, engineers conducted a simulation to test a new type of data center designed for electricity flexibility. They recreated a high-demand scenario on the UK's grid from a 2020 Euro tournament soccer match, where a surge in electric kettle use caused a substantial increase in power demand. An AI program then instructed a simulated London data center to decrease power to some of its chips, demonstrating how such a facility could help balance supply and demand.
This software, called Conductor, is the product of Washington, DC-based firm Emerald AI. It is set to be deployed in a new facility in Virginia’s Data Center Alley, where it will connect to the live grid. Conductor will manage the data center's power usage, reducing it during demand spikes while ensuring critical operations continue. Emerald's partners in this venture include Nvidia and Digital Realty, who describe it as one of the world's first 'power-flexible AI factories.'
The development of power-flexible data centers could mitigate a significant bottleneck in bringing new facilities online: the lengthy process of approving, constructing, and connecting new power plants. For example, PJM, a major US grid operator in Virginia, reportedly needs eight years to integrate new generation, according to RMI, an energy research and advocacy group. Nvidia's head of sustainability, Josh Parker, stated that "AI factory flexibility is the bridge between the incredible demand for AI and the immediate limitations of our energy grid."
Beyond speed, this flexibility also addresses public concerns regarding data centers' high electricity draw, noise, and environmental impact. In 2025, projects totaling over $150 billion were stalled, according to Data Center Watch, due to public opposition and legislative actions. Policymakers are responding with potential bans in several states and federal proposals like the bipartisan GRID Act, which suggests disconnecting new data centers from public grids entirely. By limiting power draw during peak times, data centers might avoid the need for extensive new infrastructure upgrades, utilizing the existing grid more efficiently.
According to MIT Technology Review, this approach seeks to optimize the current electrical system, which operates near full capacity only during limited periods each year.
